Voters Rejected Data Centers in 40 of 41 Votes. The Bill Arrived in the Semiconductor Index.

On Wednesday the Dow Jones Industrial Average rose 1,569.40 points, or 3.03 percent, to 53,356.12, its best session since 9 April 2025. The Philadelphia Semiconductor Index fell 3.1 percent to a six-month low in the same hours.

Both moves were rational, which is the uncomfortable bit.

The Dow was not rallying on artificial intelligence. It was rallying on the removal of an unknown. Democrats took both chambers of Congress, which means divided government and less legislation, and oil fell more than 8 percent on bets that the Iran war might end sooner. Microsoft rose 6.7 percent, Amazon 6.1 percent, Alphabet 3.2 percent and Meta 1.4 percent, according to Reuters' closing numbers. The companies buying compute were treated as the winners.

The companies selling it were not. Hut 8 fell 23 percent, IREN 11 percent, Core Scientific 10 percent and Applied Digital 6.7 percent. Nvidia fell 3.1 percent. CoreWeave's unsecured debt due in 2031 traded as low as 79.5 cents on the dollar, with its 2030 and 2032 notes yielding above 15 percent, Bloomberg reported.

One election, one session, two opposite verdicts on the same industry. The split is not a mood. It is a repricing of who carries the risk of the buildout, and it happened because 41 local ballot measures told investors something the federal result did not.

What the ballot actually decided

Ballotpedia counted 41 data center measures across 10 states. By CNBC's tally and Scientific American's, voters sided against the industry in every one but a single contest in Harwood, North Dakota, where about three in four voters rejected a proposal that would have added extra review for industrial projects, data centers included. Ohio alone had 21 of the 41. Granville's ban passed with 82 percent. Morgan County, Illinois, voted to restrict with 88 percent.

The counts differ by outlet, and the difference is worth understanding rather than averaging. The Washington Post reported at least 38 jurisdictions. WMUR reported 45 measures. CNBC and Scientific American both used Ballotpedia's 41. Measures, jurisdictions and certified results are three different things, and several counts were still pending on Wednesday. The direction is not in dispute. Voters were asked, in effect, whether they wanted a data center near them, and the answer was no almost everywhere it was asked.

The exit polling is blunt. Seventy-six percent of voters opposed building a data center in their own community, and 57 percent said AI will do more to hurt society than help, according to the NBC News exit poll. A September UMass Amherst poll found 76 percent of Democrats and 52 percent of Republicans opposed construction in their communities. Fifty-six percent told CNBC's voter poll that AI and data centers were a crucial or major factor in their vote.

The distinction matters for how the result should be read. Nobody voted on a federal AI policy. They voted on permits, water, noise, power bills and whether a county commission would say yes. The AI industry lost a planning fight, not a philosophical one, and it lost it at the level where projects are actually approved.

Austin and Columbus are the repricing events

The market did not fall because 40 local votes went the wrong way. It fell because two governorships changed hands in states where the capacity is going.

In Texas, Gina Hinojosa beat Greg Abbott. She is the first Democrat elected governor of the state since Ann Richards in 1990. Her campaign promise was specific. "When I am governor I will issue a moratorium ... on all data center approvals until laws can be passed by the Legislature to protect Texans," she said in a video posted on 22 October, as reported by the Texas Tribune.

The mechanism is less dramatic than the headline. Abbott had already stopped the machinery. On 3 August he halted new data center connections to the state grid and ordered an audit of the queue; ERCOT was tracking more than 400 gigawatts of data center projects at the time, roughly what it would take to power the entire US South, Reuters reported. He then directed the environmental regulator to issue no data center permits while the audit ran. Hinojosa does not need a new statute to keep that freeze in place. She needs one to make it permanent, and to end the sales tax exemption that costs the state billions a year, according to the Texas Tribune.

Gordon Johnson of GLJ Research put roughly 13 gigawatts of Texas capacity immediately exposed. Projects with "power, permits and steel keep going," he wrote in a note Bloomberg reported; those with power but no state environmental permits slip 12 to 18 months.

The exposure is large because Texas is large. Jefferies put the state at roughly 26 percent of planned US data center capacity, nearly three times Virginia's pipeline, and estimated that a moratorium could put 13.3 gigawatts of announced projects at risk, as Fast Company reported. Axios put the state at more than 41 percent of the capacity expected to be added in 2028. Those two numbers are not in conflict. One is a share of everything planned, the other a share of what was supposed to come online in a single year, and the second is the one that hits an order book.

Ohio flipped too. Amy Acton beat Vivek Ramaswamy after campaigning on a conditional moratorium that would require data centers to cover 100 percent of utility costs, CNBC reported. She will govern with a Republican legislature, which limits what she can legislate but not what her administration can permit. Georgia's governor's race flipped as well. Texas, Arizona, Georgia and Ohio, four of the largest state markets for the buildout, now have governors-elect who campaigned on restricting it.

Why the same news lifted Microsoft and cut Hut 8

The asymmetry is simple once you write down who owes what.

A hyperscaler spending hundreds of billions a year on infrastructure would quite like the pace to slow. Less urgency means better prices, more time to fix utilization, and less pressure to buy memory at whatever it costs this month. Microsoft, Amazon, Alphabet and Meta all have cash flows that survive a delay. Three of the four have already turned free cash flow negative funding the buildout, which is precisely why a slower schedule reads as relief rather than threat.

A supplier or a neocloud has the opposite exposure. Its revenue is the buildout. Its debt is sized against a completion date. If the date moves, the interest does not.

"If you push out the pace or change the pace of the data center buildout, you begin to change the fundamental trajectory for many of those suppliers," Scott Chronert, head of US equity strategy at Citi Research, told Reuters.

The credit market reached that conclusion before the equity market did. CoreWeave's five-year credit default swaps have traded around or above 1,000 basis points since mid-October, the level S&P Global's Gavan Nolan describes as distress, and its 2031 bonds were already yielding above 15 percent before the election. Wednesday's move to 79.5 cents was an acceleration, not a discovery. The people who own the paper have been pricing permitting risk for a month.

The number that makes this a 2028 problem

Data center construction has a long fuse. A measure passed in November 2026 does not stop a building that is already framed. Analysts quoted by Ars Technica expect the votes to bite on projects planned for 2028, not on what is under construction or already approved through 2027.

That is the uncomfortable part. Axios projected that nearly two-thirds of 2028 US data center capacity could face delays of the kind Hinojosa is threatening in Texas. Bloomberg estimated that half of US data centers are now more vulnerable to scrutiny. These are forecasts, not outcomes, and forecasts from banks have a habit of describing the direction of a negotiation rather than its result. But they describe the year the current order books are built around.

The demand side has not disappeared. Gartner's estimate that the AI boom is nearly $5 trillion short of paying for itself, reported by Fortune, is the number that makes the political risk expensive rather than merely annoying. If the revenue arrives later than the depreciation schedule, then a two-year permitting delay is not a scheduling problem. It is a financing problem for whoever is holding the asset.

The industry's counter-argument, and where it holds

The pushback arrived within hours, and part of it is correct.

"Let's not confuse an anti-Trump wave with an anti-tech mandate," said Adam Kovacevich, chief executive of the trade group Chamber of Progress. He argued that Democratic gubernatorial candidates who opposed data center bans won at a higher rate than those who supported them, and that voters wanted change on the economy rather than a crackdown on innovation.

It is a real observation about candidate-level results. It is also not a rebuttal to 76 percent. A voter can disapprove of Trump, want cheaper groceries and still not want a hyperscale facility at the end of the road. Alejandra Montoya-Boyer of The Leadership Conference's Center for Civil Rights and Technology put the other side plainly to Ars: "The 2026 midterms have been the first true AI election in the US."

The structural limits on what follows are just as real. Democrats do not have a filibuster-proof Senate. Trump has a veto. A governor controls permits, not construction, and the tax exemption that makes Texas attractive belongs to a legislature that does not convene until 12 January, a week before Hinojosa takes office. PIMCO's Libby Cantrill expects "a lot more sand in the gears" even if no statewide moratorium survives, with more time for approvals and more concessions from developers. The base case is delay and cost, not cancellation.

Which is why the sell-off was concentrated where it was. A delay costs the operator carrying costs and costs the supplier a quarter of revenue. It costs the hyperscaler an option. The market repriced the two sides accordingly, and it did so in a single session, because the information arrived in a single night.

What gets repriced first: time

The useful question for anyone selling into this buildout is not whether the AI boom is real. It is which parts of your 2027 and 2028 plan depend on a permit that has not been issued.

Three things are worth putting in writing this quarter. Which purchase orders carry a delivery date that assumes a site is energized on schedule, and what happens contractually if it is not. How much of the forecast is anchored to Texas, Georgia, Ohio or Arizona, the four states where the political risk just changed. And whether the customer's commitment survives a change in state law, or whether it is a letter of intent with a date on it.

The companies that get this right will not avoid the delay. They will know which projects are theirs to lose. It is a sourcing and planning problem before it is a demand problem, and it is the kind of problem that shows up in a forecast review rather than a news cycle.

At DMC we work with hardware companies on exactly that: demand scenarios that survive a policy change, supplier commitments with cancellation windows you can live with, and production ramp plans that do not assume a 2028 number stays a 2028 number. If your 2027 plan is built on a 2028 forecast, let's talk.